Identifying Variables Responsible for Data not Missing at Random

When data are not missing at random (NMAR), maximum likelihood (ML) procedure will not generate consistent parameter estimates unless the missing data mechanism is correctly modeled. Understanding NMAR mechanism in a data set would allow one to better use the ML methodology. A survey or questionnaire may contain many items; certain items may be responsible for NMAR values in other items. The paper develops statistical procedures to identify the responsible items. By comparing ML estimates (MLE), statistics are developed to test whether the MLEs are changed when excluding items. The items that cause a significant change of the MLEs are responsible for the NMAR mechanism. Normal distribution is used for obtaining the MLEs; a sandwich-type covariance matrix is used to account for distribution violations. The class of nonnormal distributions within which the procedure is valid is provided. Both saturated and structural models are considered. Effect sizes are also defined and studied. The results indicate that more missing data in a sample does not necessarily imply more significant test statistics due to smaller effect sizes. Knowing the true population means and covariances or the parameter values in structural equation models may not make things easier either. The research was supported by NSF grant DMS04-37167, the James McKeen Cattell Fund.     

Estimation of the Nonlinear Random Coefficient Model when Some Random Effects Are Separable

A method is presented for marginal maximum likelihood estimation of the nonlinear random coefficient model when the response function has some linear parameters. This is done by writing the marginal distribution of the repeated measures as a conditional distribution of the response given the nonlinear random effects. The resulting distribution then requires an integral equation that is of dimension equal to the number of nonlinear terms. For nonlinear functions that have linear coefficients, the improvement in computational speed and accuracy using the new algorithm can be dramatic. An illustration of the method with repeated measures data from a learning experiment is presented.     

SINGLE‐SAMPLE DISCRIMINATION OF DIFFERENT SCHEDULES' REINFORCED INTERRESPONSE TIMES

Food‐deprived rats in Experiment 1 responded to one of two tandem schedules that were, with equal probability, associated with a sample lever. The tandem schedules' initial links were different random‐interval schedules. Their values were adjusted to approximate equality in time to completing each tandem schedule's response requirements. The tandem schedules differed in their terminal links: One reinforced short interresponse times; the other reinforced long ones. Tandem‐schedule completion presented two comparison levers, one of which was associated with each tandem schedule. Pressing the lever associated with the sample‐lever tandem schedule produced a food pellet. Pressing the other produced a blackout. The difference between terminal‐link reinforced interresponse times varied across 10‐trial blocks within a session. Conditional‐discrimination accuracy increased with the size of the temporal difference between terminal‐link reinforced interresponse times. In Experiment 2, one tandem schedule was replaced by a random ratio, while the comparison schedule was either a tandem schedule that only reinforced long interresponse times or a random‐interval schedule. Again, conditional‐discrimination accuracy increased with the temporal difference between the two schedules' reinforced interresponse times. Most rats mastered the discrimination between random ratio and random interval, showing that the interresponse times reinforced by these schedules can serve to discriminate between these schedules.     

Human operant performance: Sensitivity and pseudosensitivity to contingencies

Undergraduates' button presses occasionally produced points exchangeable for money. Left and right buttons were initially correlated with multiple random-ratio (RR) and random-interval (RI) components, respectively. During interruptions of the multiple schedule, students filled out sentence-completion guess sheets describing the schedules, and points were contingent upon the accuracy of guesses. To test for sensitivity to schedule contingencies, schedule components were repeatedly reversed between the two buttons. Pressing rates were consistently higher in ratio than in interval components even when feedback for guesses was discontinued, demonstrating sensitivity to the difference between ratio and interval contingencies. The question was whether this sensitivity was based directly on the contingencies or whether it was rule-governed. For two students, when multiple RR RI schedules were changed to multiple RI RI schedules, rates became low in both components of the multiple RI RI schedule; however, subsequent prevention of point deliveries for the first few responses in any component produced high rates in that component. For a third student, response rates became higher in the RI component that provided the lower rate of reinforcement. In each case, performance was inconsistent with typical effects of the respective schedules with nonhuman organisms; it was therefore plausible to conclude that the apparently contingency-governed performances were instead rule-governed.     

Measuring misinformation in repeat trial pick 1 of 2 tests

An extension is described to a product testing model to account for misinformation among subjects. A misinformed subject is one who associates the taste of product A with product B and vice-versa; thus, the subject would tend to perform incorrectly on pick 1 of 2 tests. A likelihood ratio test for the presence of misinformation is described. The model is applied to a data set, and misinformation is found to exist. Biases due to model misspecificationand other implications for product testing are discussed.The first author is currently on leave from Carnegie Mellon University.     

Round-robin analysis of social interaction: Exact and estimated standard errors

Kenny has proposed a variance-components model for dyadic social interaction. His Social Relations model estimates variances and covariances from a round-robin of two-person interactions. The current paper presents a matrix formulation of the Social Relations model. It uses the formulation to derive exact and estimated standard errors for round-robin estimates of Social Relations parameters.We are grateful to George Gilbert for comments on an earlier draft.     

Maximum likelihood estimation of item response parameters when some responses are omitted

A theoretical model is given for dealing with omitted responses. Two special cases are investigated.This work was supported in part by contract N00014-80-C-0402, project designation NR 150-453 between the Office of Naval Research and Educational Testing Service. Reproduction in whole or in part is permitted for any purpose of the United States Government.     

Trend in correlated proportions

A random effects probit model is developed for the case in which the same units are sampled repeatedly at each level of an independent variable. Because the observed proportions may be correlated under these conditions, estimating their trend with respect to the independent variable is no longer a standard problem for probit, logit or loglinear analysis. Using a qualitative analogue of a random regressions model, we employ instead marginal maximum likelihood to estimate the average latent trend line. Likelihood ratio tests of the hypothesis of no trend in the average line, and the hypothesis of no differences in average trend lines between experimental treatments, are proposed. We illustrate the model both with simulated data and with observed data from a clinical experiment in which psychiatric patients on two drug therapies are rated on five occasions for the presence or absence of symptoms.Supported by a grant from the MacArthur Foundation and National Science Foundation Grant BNS85-11774.The authors are indebted to James Heckman for calling our attention to the Clark algorithm.     

A comparison of response patterns on fixed-, variable-, and random-ratio schedules

The behavior of individual pigeons on fixed-ratio, variable-ratio, and random-ratio schedules was examined. Within each type of ratio schedule the size of the ratio was varied in an irregular sequence. At various ratio sizes (5, 10, 40, 80) no differences were found among overall response rates (postreinforcement pause plus running response rate) as a function of ratio type. This similarity in overall response rates held despite noticeable differences in the microstructure of performance both within and across subjects; the primary performance difference on the three types of ratio schedules was the relatively longer postreinforcement pause duration on the fixed-ratio schedule. We concluded that the gross temporal characteristics of performance determined by the relative weightings of the postreinforcement pause and running response rate were primarily controlled by the type of ratio schedule (fixed, variable, or random), whereas the overall rate of responding was controlled by the size of the ratio.     

Shy teens and their peers: Shyness in respect to basic personality traits and social relations

The main purpose of this paper was to examine shyness in teenagers from two perspectives: in terms of its relations with basic personality traits and in terms of its influence on the processes that occur in the social networks of high school students. First, we found that shyness was negatively predicted by extraversion and positively by neuroticism. Second, using exponential random graph models we demonstrated that shyness across network effects was similar to reversed extraversion (introversion): Both negatively predicted the number of outgoing relations, whereas they did not affect the number of incoming relations. We discuss the issue of locating shyness in the space of personality traits, supporting the relevance of ascribing it to introversion.